Scoring Pennsylvania

Michal Migurski
Mar 1, 2018 · 4 min read

On February 13 we launched into the whirlwind of Pennsylvania’s disputed congressional plan. Our mission is to help citizens evaluate new redistricting plans for partisan skew and to elevate the conversation on partisan gerrymandering with new historical data for state and U.S. congressional district plans. It’s been an exciting couple of weeks. What did we learn, and how are we doing at reaching our goals?

Every district plan is designed to do a job: apportioning voters into Congressional districts so they can elect members to the U.S. House, typically following traditional redistricting criteria like compactness, contiguity, and preservation of communities and incumbents. Since 2010, we’ve seen the emergence of detailed computer models that allow for the creation of unfair, skewed maps that still conform to traditional criteria:

It’s also increasingly clear that the finely grained data available to map-drawers, which they can manipulate via sophisticated software, has made “traditional redistricting principles” less effective in combating gerrymanders. The Pennsylvania case in the news right now provides a great illustration: After the state Supreme Court told the legislature to come up with a new congressional map that was less partisan and also less disruptive of traditional redistricting principles, the solons promptly came up with a map that looked a lot neater and nicer but was just as partisan as the original. (Ed Kilgore, New York Magazine, Feb 20, 2018)

We are putting that same finely-grained data and sophisticated software to work combatting new partisan gerrymanders. Our predictive model of Pennsylvania, built by Eric McGhee from 2016 precinct-level election results and demographics, helped Ruth Greenwood rapidly score a cluster of maps all released during one day. In the past, Ruth and other experts have devoted multiple days to scoring plans and publishing results (such as this report on North Carolina from last year), often under enormous time pressure to meet legal deadlines. With our scoring feature for uploaded plans, this same analysis can take just minutes.

Seven different plans were published on February 15, from a spectrum of political and other interest groups. Ordinarily, evaluating such plans would require specialized knowledge, data, and time (the New York Times district-by-district analysis by Nate Cohn, Matthew Bloch, and Kevin Quealy is an excellent example). Here, we were able to rapidly compare the proposals in just a few minutes. Nicholas Stephanopoulos summarized the results in his Election Law Blog post: “it is not enough to say that we want an aesthetically attractive map; we also have to specify what kind of aesthetically attractive map we want: one that is skewed in favor of a party or one that is symmetric.” Four days later, we followed up with an instant analysis of the Pennsylvania Supreme Court’s remedial plan. Good news: it’s less skewed and more competitive, offering voters a more meaningful choice when they head to the polls this November!

Summary of PlanScore’s evaluation of the PA remedial plan

Our scoring model doesn’t take incumbency into account, instead relying on 2016 election results and demographic data. Based on feedback to our work, we’re developing a way to incorporate incumbents into the model. This is an advanced feature, but we think it will help improve our predictions for plans where incumbent candidates of either party are known. If you can tell us who’s running for office and you know where they live, you will get a better result.

Historical data for Pennsylvania

The other half of offers historical data based on Simon Jackman’s work with Eric and Nicholas. Our friends at GreenInfo Network turned five decades of data for national and state plans into the beautiful, explorable data visualization on our home page and fifty state pages. Nicholas highlighted a few tidbits of history that we think will become increasingly interesting as we approach national redistricting in 2020:

  • Why were Republicans upset enough about California’s 1982 congressional plan that they launched a (successful) statewide referendum to get rid of it? Because the plan was one of the most pro-Democratic on record, with a double-digit pro-Democratic efficiency gap.
  • What happened after a court replaced Georgia’s 2002 state house plan — the subject of two Supreme Court cases — with a map of its own creation? The old plan’s large Democratic skew gave way to significantly greater partisan balance in the new map.
  • And why were Texas Republicans so keen to re-redistrict the state’s congressional plan in 2004? Because the plan sharply benefited Democrats, and by redrawing it Republicans were able to shift its efficiency gap more than ten points in their favor.

The historical partisan symmetry scores for Pennsylvania tell a similar story. The very fair (on all three metrics) 1992 plan gave way to a historically skewed 2012 plan, making the current out-of-cycle redistricting remedy necessary to ensure a fair map.

U.S. House results under Pennsylvania’s fair 1992 plan (left) and its disputed 2012 plan (right)

All of this will provide critical context for the next three years of map drawing, culminating in the results of the 2020 Census. Along the way, we’ll see a Supreme Court ruling on Wisconsin’s map in Gill v. Whitford, a decision on Maryland and North Carolina, and further debate about the effects of partisan map-drawing. We’re excited to extend our model to all fifty states, and for this data to inform the next three years of redistricting conversation!


Measuring partisan gerrymandering